Seventeen data center projects. Zero megawatts disclosed.
That is the entire quantitative payload of a report now circulating through crypto media โ a finding attributed to the Center for Western Priorities, claiming that a cluster of hyperscale data center proposals has landed on federally managed land across the American West. Seventeen sites. Public land. The word "proposed" doing all the heavy lifting.
I want you to sit with the asymmetry for a moment. This week I traced a wallet cluster moving $41.7 million across eleven hops in under nine hours, and I could tell you the entry exchange, the bridge used, the exact block heights, and the gas premium paid at each step. I can reconstruct a laundering chain to the dollar. I cannot tell you whether one of these seventeen projects is a 40MW edge node or a 900MW training campus, because nobody published a number. Not the NGO. Not the developer. Not the federal agency holding the lease file.
A pipeline without megawatts is not a pipeline. It is a rumor with a filing fee attached. And in a bull market, rumors with filing fees get priced as if they were steel on the ground. That gap โ between the cost of an option and the cost of construction โ is where the entire story actually lives.
Context: what the report says, and what three-hand information can carry
The sourcing chain here matters more than the headline, so let me lay it out clinically. An advocacy NGO publishes a finding. A journalist restates it. A crypto outlet re-publishes the restatement. What reaches you is a third-generation copy, and every generation sheds the load-bearing details โ project names, counties, acreage, developer identity, megawatt class, interconnection status, environmental review stage. The original document may well contain those. The version in circulation contains none of them.
I have seen this pattern before. Where early ICO ghosts still haunt the ledger, they haunt it the same way: a whitepaper with a token supply and no vesting schedule, a Telegram with 40,000 members and no GitHub commit in six weeks. The tell was never the ambition. The tell was always the missing field โ the one number that a real operator would have no reason to hide and every reason to publish.
So let me be precise about what this report is and is not.
It is a signal of a policy-resource collision that the market has badly under-priced: federal land as the substrate for AI-era compute. That is a genuinely under-discussed intersection, and the report deserves credit for surfacing it even in degraded form.
It is not evidence of anything being built. The industry base rate for "proposed" converting to "operational" in data center development sits under 20%, and that base rate was computed in a benign interest-rate environment with functioning transformer supply chains. Neither condition holds today.
Here is the arithmetic the report omitted, and which I ran myself. If those seventeen projects hold a median of roughly 300MW of IT load each โ which is the current convention for an AI-capable campus, with the range running from about 100MW at the small end to well past 1GW at the hyperscale end โ the aggregate pipeline is on the order of 5 gigawatts. At industry-standard capital intensity of $8โ12 million per megawatt, that implies $40โ60 billion in capex, and somewhere between 40 and 60 terawatt-hours of annual consumption once energized. For scale: that is a low-double-digit percentage of total United States data center electricity demand as of 2023, arriving from seventeen parcels of land that may or may not ever break ground.
That is the number that should have been in the headline. It wasn't, because nobody on the reporting chain had it. A reader who cannot compute the megawatt arithmetic cannot distinguish a national infrastructure shift from a press release with a map graphic.
Core: the actual mechanics of why public land, and why now
The lazy read is that federal land is cheap land, and that developers are chasing a discount. That read is wrong, and the wrongness is instructive.
Land is typically 1% to 3% of total project capex in a data center build. If you are spending $3 billion on a 300MW campus, the dirt is a rounding error against the switchgear, the chillers, the transformers, the liquid cooling loops, and the GPUs. Nobody structures a multi-billion-dollar infrastructure program around a rounding error. So the discount thesis fails on its own math.
What public land actually offers is something far more valuable and far less discussed: a single counterparty that controls the ground, the transmission adjacency, and the permitting authority at the same time. That is the real product. In the private market, those three things are held by three different parties with three different clocks and three different veto rights. On federal land in the West, they frequently converge โ the Bureau of Land Management holds surface rights, the Western Area Power Administration operates transmission corridors nearby, and in some cases the acreage sits adjacent to federal hydroelectric infrastructure on the Colorado River system. One landlord. One grid neighbor. One approval chain.
That convergence is worth more than free dirt, because the binding constraint in this industry is not capital and it is not silicon. It is time-to-power.
A 300MW campus that energizes eighteen months earlier than a competitor does not earn 18 months of extra revenue; it earns 18 months of contracted revenue at whatever the prevailing wholesale rate was when the market was tightest, locked into a 10-to-15-year lease. On a $150/kW/month wholesale colocation rate, a single 300MW campus bills roughly $45 million per month at stabilization. Eighteen months of that is north of $800 million of pulled-forward cash flow. Against a $3 billion build, that is not a marginal advantage. It is the difference between a fundable project and an unfundable one.
Which reframes the entire report. The Center for Western Priorities is worried about resource occupation โ water, land, viewshed, wildlife corridor. Those are legitimate concerns. But the economic substance of the public land question is not conservation versus development; it is queue position versus queue position. The land is a means. Speed is the asset.
And speed is exactly what is being rationed. Reported lead times for large power transformers have run past 120 weeks in recent cycles. Heavy-duty gas turbine slots are effectively sold out into the late 2020s. Interconnection queues in the major Western grid regions remain backlogged despite post-Order-2023 reforms that added deposits and readiness milestones precisely to flush out speculative placeholders. Every one of those constraints is a clock, and every clock favors whoever already holds a position.
This is where the analytical lens from on-chain forensics becomes unexpectedly useful. A queue position is a financial instrument, and it has the same structure as an option: a strike price (study deposits, readiness payments, interconnection agreements), an expiry (queue reform deadlines and milestone penalties), and a payoff that is highly sensitive to volatility in the underlying โ in this case, AI compute demand. Options get priced even when they never exercise. In fact, options are most actively priced in exactly the regimes where exercise probability is lowest and variance is highest.
So when seventeen "proposed" projects appear on federal land, the correct interpretation is not "seventeen data centers are coming." The correct interpretation is "seventeen real options on Western electrons have been written, and the market is quoting them as if the underlying were already delivered."
There is a negative externality hiding in that framing, and it is the part policymakers keep missing. A speculative queue position is not neutral. It occupies a slot that a serious developer cannot use. It forces the grid operator to run studies on a project that will never energize. It extends the backlog for everyone behind it. If even half of those seventeen parcels are land-banking exercises โ and the base rate suggests more than half โ then the surge in federal land proposals is partly causing the scarcity it claims to be responding to. The pipeline is inflating its own constraint.
Now the part that connects this to the ledger you and I actually trade against.
The megawatt has quietly become a multi-use asset, and the market has already repriced it. Bitcoin miners discovered this before the AI funds did. Consider the per-megawatt economics. A 1MW mining deployment running modern rigs at roughly 20 joules per terahash yields about 50 petahash. At a post-halving hashprice near $50 per petahash per day, that is roughly $2,500 of daily gross revenue; subtract power at five cents a kilowatt-hour โ about $1,200 a day โ and you are clearing somewhere around $1,300 per megawatt per day, before depreciation and before the next difficulty adjustment eats your lunch. Now price the same megawatt as an AI hosting slot at $150/kW/month wholesale: $150,000 a month, or roughly $5,000 a day.
That is a three-to-fourfold repricing of the identical physical input. You do not need a model to see what happens next. The publicly disclosed hosting conversions โ the Core Scientific and CoreWeave arrangement, Hut 8, IREN, TeraWulf, Cipher โ are not pivots away from crypto. They are the same trade expressed in a different denominator. Firm power and a signed interconnection agreement became the scarcest asset in the digital economy, and the miners happened to be holding it.
The query I run to watch this is unglamorous. The shape of it looks like this:
SELECT
entity,
SUM(usd_value) AS outbound_usd_90d,
COUNT(DISTINCT counterparty) AS payees,
MIN(block_time) AS first_seen
FROM wallet_transfers
WHERE chain = 'ethereum'
AND direction = 'outbound'
AND block_time >= NOW() - INTERVAL '90 days'
AND entity_label IN ('Core Scientific','Hut 8','IREN','TeraWulf','Cipher')
GROUP BY entity
ORDER BY outbound_usd_90d DESC;I am not going to hand you a fabricated result set. What I will tell you is what the pattern looks like when it's real: capital expenditure announcements lead on-chain settlement by one to two quarters, and the tell is not the headline number โ it is the diversification of payees. A miner paying a single equipment vendor is buying rigs. A miner paying forty distinct counterparties across construction, substation engineering, and cooling contractors is building a shell. That is the signature of a hosting conversion, and it shows up in wallet behavior before it shows up in a press release.
The same discipline applies to the federal land story, and its absence is the whole problem. We can watch a mining treasury move to the dollar. We cannot watch a 300MW campus because nobody published the megawatt. The information asymmetry is not a gap in the data; it is a choice by the disclosing parties.
And here is where I have to puncture a narrative that the bull market is currently pricing at a premium. There is a faction arguing that this infrastructure pipeline will be financialized โ tokenized land rights, tokenized power purchase agreements, tokenized interconnection capacity. Real-world-asset protocols have been running that story for three years now, and the story has consistently outrun the plumbing. An interconnection queue position is not a bearer asset. It is a contractual status with a named counterparty, a regulatory body, and a milestone schedule. It cannot be meaningfully transferred without the grid operator's consent, which means it cannot be meaningfully tokenized without reintroducing exactly the trusted intermediary the token was supposed to eliminate. You can wrap it in a legal entity and sell shares of the entity. That is called a fund. It is not innovation; it is a fund with extra steps and a worse audit trail.
The institutional demand side is real โ my own work mapping data flows between decentralized compute networks and AI training corpora found a meaningful share of high-value training data originating from verifiable on-chain sources. But demand for verifiable data is not demand for tokenized dirt. Conflating the two is how the RWA sector keeps generating three-year storytelling exercises while the actual scarce asset โ a signed, firm, interruptible-free power contract โ sits in a filing cabinet in a county office.
Contrarian: the surge may be a financing artifact, not a construction program
Here is the angle almost nobody is running, and it is the one I would stake capital on.
Correlation is not causation, and a cluster of proposals is not a cluster of buildings. The visible surge in federal land data center filings coincides almost perfectly with two other curves: the collapse of mining margins after the 2024 halving, and the peak of AI infrastructure fundraising. Both of those curves create a powerful incentive to announce rather than to build. Announcements unlock capital. Announcements justify land banking. Announcements let a developer with a stranded 200MW interconnection position mark it at AI-hosting valuations instead of mining valuations.
Which means the most probable composition of those seventeen projects is not seventeen data centers. It is something closer to three or four genuinely pre-leased hyperscale builds with named tenants and firm power, six to eight speculative shells with an interconnection deposit and a site plan, and a tail of pure optionality โ acreage held against a future where somebody else's demand curve does the work.
The second contrarian point is harder for the AI-maximalists to hear. The constraint was never land, and public land was never the solution. Firm power is the constraint. And the fastest firm power in America right now is not being built on federal acreage adjacent to a Colorado River dam that requires a multi-year environmental review. It is being built behind the meter, on private land, next to gas pipelines, in Texas and the PJM footprint, with turbine orders placed eighteen months ago by people who understood the bottleneck before it became a headline. Federal land offers a beautiful landlord story and a brutal permitting timeline. Those two facts do not cancel. They compete, and the timeline usually wins.
The data doesn't negotiate. Seventeen proposals, zero megawatts, no disclosed tenants โ that is a description of a marketing surface, not a supply chain.
Takeaway: what to watch next quarter
Ignore the seventeenth project. Watch the ratio. The single most informative metric in this entire story is the pre-leased-to-speculative ratio inside the federal land pipeline, and it will surface through county records, interconnection milestone filings, and utility dockets long before it surfaces in a press release. If that ratio deteriorates, the "surge" is a financing event and the electrification timeline is fiction. If it holds, you are watching the early outline of a genuine GW-scale reallocation of Western power โ and the miners holding firm capacity are the ones who get paid twice.
Precision in chaos is the only true advantage. The chaos here is a seventeen-line list with no numbers attached. The precision is knowing that the numbers were always the only thing that mattered, and that their absence was never an accident.